1 /*
2  * TO DO:
3  *	- gc of file systems (not going to do just yet?)
4  *	- statistics file
5  *	- configure on amsterdam
6  */
7 
8 #include <u.h>
9 #include <libc.h>
10 #include <bio.h>
11 #include <ip.h>
12 #include <thread.h>
13 #include <libsec.h>
14 #include <sunrpc.h>
15 #include <nfs3.h>
16 #include <diskfs.h>
17 #include <venti.h>
18 #include "nfs3srv.h"
19 
20 #define trace if(!tracecalls){}else print
21 
22 typedef struct Ipokay Ipokay;
23 typedef struct Config Config;
24 typedef struct Ctree Ctree;
25 typedef struct Cnode Cnode;
26 
27 struct Ipokay
28 {
29 	int okay;
30 	uchar ip[IPaddrlen];
31 	uchar mask[IPaddrlen];
32 };
33 
34 struct Config
35 {
36 	Ipokay *ok;
37 	uint nok;
38 	ulong mtime;
39 	Ctree *ctree;
40 };
41 
42 char 		*addr;
43 int			blocksize;
44 int			cachesize;
45 Config		config;
46 char			*configfile;
47 int			encryptedhandles = 1;
48 Channel		*nfschan;
49 Channel		*mountchan;
50 Channel		*timerchan;
51 Nfs3Handle	root;
52 SunSrv		*srv;
53 int			tracecalls;
54 VtCache		*vcache;
55 VtConn		*z;
56 
57 void			cryptinit(void);
58 void			timerthread(void*);
59 void			timerproc(void*);
60 
61 extern	void			handleunparse(Fsys*, Nfs3Handle*, Nfs3Handle*, int);
62 extern	Nfs3Status	handleparse(Nfs3Handle*, Fsys**, Nfs3Handle*, int);
63 
64 Nfs3Status	logread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
65 Nfs3Status	refreshdiskread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
66 Nfs3Status	refreshconfigread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
67 
68 int readconfigfile(Config *cp);
69 void setrootfid(void);
70 int ipokay(uchar *ip, ushort port);
71 
72 u64int	unittoull(char*);
73 
74 void
75 usage(void)
76 {
77 	fprint(2, "usage: vnfs [-LLRVir] [-a addr] [-b blocksize] [-c cachesize] configfile\n");
78 	threadexitsall("usage");
79 }
80 
81 void
82 threadmain(int argc, char **argv)
83 {
84 	fmtinstall('B', sunrpcfmt);
85 	fmtinstall('C', suncallfmt);
86 	fmtinstall('F', vtfcallfmt);
87 	fmtinstall('H', encodefmt);
88 	fmtinstall('I', eipfmt);
89 	fmtinstall('V', vtscorefmt);
90 	sunfmtinstall(&nfs3prog);
91 	sunfmtinstall(&nfsmount3prog);
92 
93 	addr = "udp!*!2049";
94 	blocksize = 8192;
95 	cachesize = 400;
96 	srv = sunsrv();
97 	srv->ipokay = ipokay;
98 	cryptinit();
99 
100 	ARGBEGIN{
101 	default:
102 		usage();
103 	case 'E':
104 		encryptedhandles = 0;
105 		break;
106 	case 'L':
107 		if(srv->localonly == 0)
108 			srv->localonly = 1;
109 		else
110 			srv->localparanoia = 1;
111 		break;
112 	case 'R':
113 		srv->chatty++;
114 		break;
115 	case 'T':
116 		tracecalls = 1;
117 		break;
118 	case 'V':
119 		if(chattyventi++)
120 			vttracelevel++;
121 		break;
122 	case 'a':
123 		addr = EARGF(usage());
124 		break;
125 	case 'b':
126 		blocksize = unittoull(EARGF(usage()));
127 		break;
128 	case 'c':
129 		cachesize = unittoull(EARGF(usage()));
130 		break;
131 	case 'i':
132 		insecure = 1;
133 		break;
134 	case 'r':
135 		srv->alwaysreject++;
136 		break;
137 	}ARGEND
138 
139 	if(argc != 1)
140 		usage();
141 
142 	if((z = vtdial(nil)) == nil)
143 		sysfatal("vtdial: %r");
144 	if(vtconnect(z) < 0)
145 		sysfatal("vtconnect: %r");
146 	if((vcache = vtcachealloc(z, blocksize*cachesize)) == nil)
147 		sysfatal("vtcache: %r");
148 
149 	configfile = argv[0];
150 	if(readconfigfile(&config) < 0)
151 		sysfatal("readConfig: %r");
152 	setrootfid();
153 
154 	nfschan = chancreate(sizeof(SunMsg*), 0);
155 	mountchan = chancreate(sizeof(SunMsg*), 0);
156 	timerchan = chancreate(sizeof(void*), 0);
157 	
158 	if(sunsrvudp(srv, addr) < 0)
159 		sysfatal("starting server: %r");
160 
161 	sunsrvthreadcreate(srv, nfs3proc, nfschan);
162 	sunsrvthreadcreate(srv, mount3proc, mountchan);
163 	sunsrvthreadcreate(srv, timerthread, nil);
164 	proccreate(timerproc, nil, 32768);
165 	
166 	sunsrvprog(srv, &nfs3prog, nfschan);
167 	sunsrvprog(srv, &nfsmount3prog, mountchan);
168 
169 	threadexits(nil);
170 }
171 
172 #define TWID64	((u64int)~(u64int)0)
173 
174 u64int
175 unittoull(char *s)
176 {
177 	char *es;
178 	u64int n;
179 
180 	if(s == nil)
181 		return TWID64;
182 	n = strtoul(s, &es, 0);
183 	if(*es == 'k' || *es == 'K'){
184 		n *= 1024;
185 		es++;
186 	}else if(*es == 'm' || *es == 'M'){
187 		n *= 1024*1024;
188 		es++;
189 	}else if(*es == 'g' || *es == 'G'){
190 		n *= 1024*1024*1024;
191 		es++;
192 	}else if(*es == 't' || *es == 'T'){
193 		n *= 1024*1024;
194 		n *= 1024*1024;
195 	}
196 	if(*es != '\0')
197 		return TWID64;
198 	return n;
199 }
200 
201 /*
202  * Handles.
203  * 
204  * We store all the state about which file a client is accessing in
205  * the handle, so that we don't have to maintain any per-client state
206  * ourselves.  In order to avoid leaking handles or letting clients 
207  * create arbitrary handles, we sign and encrypt each handle with
208  * AES using a key selected randomly when the server starts.
209  * Thus, handles cannot be used across sessions.  
210  *
211  * The decrypted handles begin with the following header:
212  *
213  *	sessid[8]		random session id chosen at start time
214  *	len[4]		length of handle that follows
215  *
216  * If we're pressed for space in the rest of the handle, we could
217  * probably reduce the amount of sessid bytes.  Note that the sessid
218  * bytes must be consistent during a run of vnfs, or else some 
219  * clients (e.g., Linux 2.4) eventually notice that successive TLookups
220  * return different handles, and they return "Stale NFS file handle"
221  * errors to system calls in response (even though we never sent
222  * that error!).
223  *
224  * Security woes aside, the fact that we have to shove everything
225  * into the handles is quite annoying.  We have to encode, in 40 bytes:
226  *
227  *	- position in the synthesized config tree
228  *	- enough of the path to do glob matching
229  *	- position in an archived file system image
230  *
231  * and the handles need to be stable across changes in the config file
232  * (though not across server restarts since encryption screws
233  * that up nicely).
234  * 
235  * We encode each of the first two as a 10-byte hash that is 
236  * the first half of a SHA1 hash.  
237  */
238 
239 enum
240 {
241 	SessidSize = 8,
242 	HeaderSize = SessidSize+4,
243 	MaxHandleSize = Nfs3MaxHandleSize - HeaderSize
244 };
245 
246 AESstate		aesstate;
247 uchar		sessid[SessidSize];
248 
249 static void
250 hencrypt(Nfs3Handle *h)
251 {
252 	uchar *p;
253 	AESstate aes;
254 
255 	/*
256 	 * root handle has special encryption - a single 0 byte - so that it
257 	 * never goes stale.
258 	 */
259 	if(h->len == root.len && memcmp(h->h, root.h, root.len) == 0){
260 		h->h[0] = 0;
261 		h->len = 1;
262 		return;
263 	}
264 
265 	if(!encryptedhandles)
266 		return;
267 
268 	if(h->len > MaxHandleSize){
269 		/* oops */
270 		fprint(2, "handle too long: %.*lH\n", h->len, h->h);
271 		memset(h->h, 'X', Nfs3MaxHandleSize);
272 		h->len = Nfs3MaxHandleSize;
273 		return;
274 	}
275 
276 	p = h->h;
277 	memmove(p+HeaderSize, p, h->len);
278 	memmove(p, sessid, SessidSize);
279 	*(u32int*)(p+SessidSize) = h->len;
280 	h->len += HeaderSize;
281 
282 	if(encryptedhandles){
283 		while(h->len < MaxHandleSize)
284 			h->h[h->len++] = 0;
285 		aes = aesstate;
286 		aesCBCencrypt(h->h, MaxHandleSize, &aes);
287 	}
288 }
289 
290 static Nfs3Status
291 hdecrypt(Nfs3Handle *h)
292 {
293 	AESstate aes;
294 	
295 	if(h->len == 1 && h->h[0] == 0){	/* single 0 byte is root */
296 		*h = root;
297 		return Nfs3Ok;
298 	}
299 
300 	if(!encryptedhandles)
301 		return Nfs3Ok;
302 
303 	if(h->len <= HeaderSize)
304 		return Nfs3ErrBadHandle;
305 	if(encryptedhandles){
306 		if(h->len != MaxHandleSize)
307 			return Nfs3ErrBadHandle;
308 		aes = aesstate;
309 		aesCBCdecrypt(h->h, h->len, &aes);
310 	}
311 	if(memcmp(h->h, sessid, SessidSize) != 0)
312 		return Nfs3ErrStale;	/* give benefit of doubt */
313 	h->len = *(u32int*)(h->h+SessidSize);
314 	if(h->len >= MaxHandleSize-HeaderSize)
315 		return Nfs3ErrBadHandle;
316 	memmove(h->h, h->h+HeaderSize, h->len);
317 	return Nfs3Ok;
318 }
319 
320 void
321 cryptinit(void)
322 {
323 	uchar key[32], ivec[AESbsize];
324 	int i;
325 	
326 	*(u32int*)sessid = truerand();
327 	for(i=0; i<nelem(key); i+=4)
328 		*(u32int*)&key[i] = truerand();
329 	for(i=0; i<nelem(ivec); i++)
330 		ivec[i] = fastrand();
331 	setupAESstate(&aesstate, key, sizeof key, ivec);
332 }
333 
334 /*
335  * Config file.
336  *
337  * The main purpose of the configuration file is to define a tree
338  * in which the archived file system images are mounted.
339  * The tree is stored as Entry structures, defined below.
340  *
341  * The configuration file also allows one to define shell-like
342  * glob expressions matching paths that are not to be displayed.
343  * The matched files or directories are shown in directory listings
344  * (could suppress these if we cared) but they cannot be opened,
345  * read, or written, and getattr returns zeroed data.
346  */
347 enum
348 {
349 	/* sizes used in handles; see nfs server below */
350 	CnodeHandleSize = 8,
351 	FsysHandleOffset = CnodeHandleSize
352 };
353 
354 /*
355  * Config file tree.
356  */
357 struct Ctree
358 {
359 	Cnode *root;
360 	Cnode *hash[1024];
361 };
362 
363 struct Cnode
364 {
365 	char *name;	/* path element */
366 	Cnode *parent;	/* in tree */
367 	Cnode *nextsib;	/* in tree */
368 	Cnode *kidlist;	/* in tree */
369 	Cnode *nexthash;	/* in hash list */
370 	
371 	Nfs3Status (*read)(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);	/* synthesized read fn */
372 
373 	uchar handle[VtScoreSize];	/* sha1(path to here) */
374 	ulong mtime;	/* mtime for this directory entry */
375 	
376 	/* fsys overlay on this node */
377 	Fsys *fsys;	/* cache of memory structure */
378 	Nfs3Handle fsyshandle;
379 	int isblackhole;	/* walking down keeps you here */
380 
381 	/*
382 	 * mount point info.
383 	 * if a mount point is inside another file system,
384 	 * the fsys and fsyshandle above have the old fs info,
385 	 * the mfsys and mfsyshandle below have the new one.
386 	 * getattrs must use the old info for consistency.
387 	 */
388 	int ismtpt;	/* whether there is an fsys mounted here */
389 	uchar fsysscore[VtScoreSize];	/* score of fsys image on venti */
390 	char *fsysimage;	/* raw disk image */
391 	Fsys *mfsys;	/* mounted file system (nil until walked) */
392 	Nfs3Handle mfsyshandle;	/* handle to root of mounted fsys */
393 	
394 	int mark;	/* gc */
395 };
396 
397 static uint
398 dumbhash(uchar *s)
399 {
400 	return (s[0]<<2)|(s[1]>>6);	/* first 10 bits */
401 }
402 
403 static Cnode*
404 mkcnode(Ctree *t, Cnode *parent, char *elem, uint elen, char *path, uint plen)
405 {
406 	uint h;
407 	Cnode *n;
408 	
409 	n = emalloc(sizeof *n + elen+1);
410 	n->name = (char*)(n+1);
411 	memmove(n->name, elem, elen);
412 	n->name[elen] = 0;
413 	n->parent = parent;
414 	if(parent){
415 		n->nextsib = parent->kidlist;
416 		parent->kidlist = n;
417 	}
418 	n->kidlist = nil;
419 	sha1((uchar*)path, plen, n->handle, nil);
420 	h = dumbhash(n->handle);
421 	n->nexthash = t->hash[h];
422 	t->hash[h] = n;
423 	
424 	return n;
425 }
426 
427 void
428 markctree(Ctree *t)
429 {
430 	int i;
431 	Cnode *n;
432 
433 	for(i=0; i<nelem(t->hash); i++)
434 		for(n=t->hash[i]; n; n=n->nexthash)
435 			if(n->name[0] != '+')
436 				n->mark = 1;
437 }
438 
439 int
440 refreshdisk(void)
441 {
442 	int i;
443 	Cnode *n;
444 	Ctree *t;
445 	
446 	t = config.ctree;
447 	for(i=0; i<nelem(t->hash); i++)
448 		for(n=t->hash[i]; n; n=n->nexthash){
449 			if(n->mfsys)
450 				disksync(n->mfsys->disk);
451 			if(n->fsys)
452 				disksync(n->fsys->disk);
453 		}
454 	return 0;
455 }
456 
457 void
458 sweepctree(Ctree *t)
459 {
460 	int i;
461 	Cnode *n;
462 
463 	/* just zero all the garbage and leave it linked into the tree */
464 	for(i=0; i<nelem(t->hash); i++){
465 		for(n=t->hash[i]; n; n=n->nexthash){
466 			if(!n->mark)
467 				continue;
468 			n->fsys = nil;
469 			free(n->fsysimage);
470 			n->fsysimage = nil;
471 			memset(n->fsysscore, 0, sizeof n->fsysscore);
472 			n->mfsys = nil;
473 			n->ismtpt = 0;
474 			memset(&n->fsyshandle, 0, sizeof n->fsyshandle);
475 			memset(&n->mfsyshandle, 0, sizeof n->mfsyshandle);
476 		}
477 	}
478 }
479 
480 static Cnode*
481 cnodewalk(Cnode *n, char *name, uint len, int markokay)
482 {
483 	Cnode *nn;
484 	
485 	for(nn=n->kidlist; nn; nn=nn->nextsib)
486 		if(strncmp(nn->name, name, len) == 0 && nn->name[len] == 0)
487 		if(!nn->mark || markokay)
488 			return nn;
489 	return nil;
490 }
491 
492 Cnode*
493 ctreewalkpath(Ctree *t, char *name, ulong createmtime)
494 {
495 	Cnode *n, *nn;
496 	char *p, *nextp;
497 	
498 	n = t->root;
499 	p = name;
500 	for(; *p; p=nextp){
501 		n->mark = 0;
502 		assert(*p == '/');
503 		p++;
504 		nextp = strchr(p, '/');
505 		if(nextp == nil)
506 			nextp = p+strlen(p);
507 		if((nn = cnodewalk(n, p, nextp-p, 1)) == nil){
508 			if(createmtime == 0)
509 				return nil;
510 			nn = mkcnode(t, n, p, nextp-p, name, nextp-name);
511 			nn->mtime = createmtime;
512 		}
513 		if(nn->mark)
514 			nn->mark = 0;
515 		n = nn;
516 	}
517 	n->mark = 0;
518 	return n;
519 }
520 
521 Ctree*
522 mkctree(void)
523 {
524 	Ctree *t;
525 	
526 	t = emalloc(sizeof *t);
527 	t->root = mkcnode(t, nil, "", 0, "", 0);
528 	
529 	ctreewalkpath(t, "/+log", time(0))->read = logread;
530 	ctreewalkpath(t, "/+refreshdisk", time(0))->read = refreshdiskread;
531 	ctreewalkpath(t, "/+refreshconfig", time(0))->read = refreshconfigread;
532 
533 	return t;
534 }
535 
536 Cnode*
537 ctreemountfsys(Ctree *t, char *path, ulong time, uchar *score, char *file)
538 {
539 	Cnode *n;
540 	
541 	if(time == 0)
542 		time = 1;
543 	n = ctreewalkpath(t, path, time);
544 	if(score){
545 		if(n->ismtpt && (n->fsysimage || memcmp(n->fsysscore, score, VtScoreSize) != 0)){
546 			free(n->fsysimage);
547 			n->fsysimage = nil;
548 			n->fsys = nil;	/* leak (might be other refs) */
549 		}
550 		memmove(n->fsysscore, score, VtScoreSize);
551 	}else{
552 		if(n->ismtpt && (n->fsysimage==nil || strcmp(n->fsysimage, file) != 0)){
553 			free(n->fsysimage);
554 			n->fsysimage = nil;
555 			n->fsys = nil;	/* leak (might be other refs) */
556 		}
557 		n->fsysimage = emalloc(strlen(file)+1);
558 		strcpy(n->fsysimage, file);
559 	}
560 	n->ismtpt = 1;
561 	return n;
562 }
563 
564 Cnode*
565 cnodebyhandle(Ctree *t, uchar *p)
566 {
567 	int h;
568 	Cnode *n;
569 	
570 	h = dumbhash(p);
571 	for(n=t->hash[h]; n; n=n->nexthash)
572 		if(memcmp(n->handle, p, CnodeHandleSize) == 0)
573 			return n;
574 	return nil;
575 }
576 
577 static int
578 parseipandmask(char *s, uchar *ip, uchar *mask)
579 {
580 	char *p, *q;
581 	
582 	p = strchr(s, '/');
583 	if(p)
584 		*p++ = 0;
585 	if(parseip(ip, s) == ~0UL)
586 		return -1;
587 	if(p == nil)
588 		memset(mask, 0xFF, IPaddrlen);
589 	else{
590 		if(isdigit((uchar)*p) && strtol(p, &q, 10)>=0 && *q==0)
591 			*--p = '/';
592 		if(parseipmask(mask, p) == ~0UL)
593 			return -1;
594 		if(*p != '/')
595 			*--p = '/';
596 	}
597 /*fprint(2, "parseipandmask %s => %I %I\n", s, ip, mask); */
598 	return 0;
599 }
600 
601 static int
602 parsetime(char *s, ulong *time)
603 {
604 	ulong x;
605 	char *p;
606 	int i;
607 	Tm tm;
608 	
609 	/* decimal integer is seconds since 1970 */
610 	x = strtoul(s, &p, 10);
611 	if(x > 0 && *p == 0){
612 		*time = x;
613 		return 0;
614 	}
615 	
616 	/* otherwise expect yyyy/mmdd/hhmm */
617 	if(strlen(s) != 14 || s[4] != '/' || s[9] != '/')
618 		return -1;
619 	for(i=0; i<4; i++)
620 		if(!isdigit((uchar)s[i]) || !isdigit((uchar)s[i+5]) || !isdigit((uchar)s[i+10]))
621 			return -1;
622 	memset(&tm, 0, sizeof tm);
623 	tm.year = atoi(s)-1900;
624 	if(tm.year < 0 || tm.year > 200)
625 		return -1;
626 	tm.mon = (s[5]-'0')*10+s[6]-'0' - 1;
627 	if(tm.mon < 0 || tm.mon > 11)
628 		return -1; 
629 	tm.mday = (s[7]-'0')*10+s[8]-'0';
630 	if(tm.mday < 0 || tm.mday > 31)
631 		return -1;
632 	tm.hour = (s[10]-'0')*10+s[11]-'0';
633 	if(tm.hour < 0 || tm.hour > 23)
634 		return -1;
635 	tm.min = (s[12]-'0')*10+s[13]-'0';
636 	if(tm.min < 0 || tm.min > 59)
637 		return -1;
638 	strcpy(tm.zone, "XXX");	/* anything but GMT */
639 if(0){
640 print("tm2sec %d/%d/%d/%d/%d\n",
641 	tm.year, tm.mon, tm.mday, tm.hour, tm.min);
642 }
643 	*time = tm2sec(&tm);
644 if(0) print("time %lud\n", *time);
645 	return 0;
646 }
647 
648 
649 int
650 readconfigfile(Config *cp)
651 {
652 	char *f[10], *image, *p, *pref, *q, *name;
653 	int nf, line;
654 	uchar scorebuf[VtScoreSize], *score;
655 	ulong time;
656 	Biobuf *b;
657 	Config c;
658 	Dir *dir;
659 
660 	name = configfile;
661 	c = *cp;
662 	if((dir = dirstat(name)) == nil)
663 		return -1;
664 	if(c.mtime == dir->mtime){
665 		free(dir);
666 		return 0;
667 	}
668 	c.mtime = dir->mtime;
669 	free(dir);
670 	if((b = Bopen(name, OREAD)) == nil)
671 		return -1;
672 	
673 	/*
674 	 * Reuse old tree, garbage collecting entries that
675 	 * are not mentioned in the new config file.
676 	 */
677 	if(c.ctree == nil)
678 		c.ctree = mkctree();
679 
680 	markctree(c.ctree);
681 	c.ok = nil;
682 	c.nok = 0;
683 	
684 	line = 0;
685 	for(; (p=Brdstr(b, '\n', 1)) != nil; free(p)){
686 		line++;
687 		if((q = strchr(p, '#')) != nil)
688 			*q = 0;
689 		nf = tokenize(p, f, nelem(f));
690 		if(nf == 0)
691 			continue;
692 		if(strcmp(f[0], "mount") == 0){
693 			if(nf != 4){
694 				werrstr("syntax error: mount /path /dev|score mtime");
695 				goto badline;
696 			}
697 			if(f[1][0] != '/'){
698 				werrstr("unrooted path %s", f[1]);
699 				goto badline;
700 			}
701 			score = nil;
702 			image = nil;
703 			if(f[2][0] == '/'){
704 				if(access(f[2], AEXIST) < 0){
705 					werrstr("image %s does not exist", f[2]);
706 					goto badline;
707 				}
708 				image = f[2];
709 			}else{
710 				if(vtparsescore(f[2], &pref, scorebuf) < 0){
711 					werrstr("bad score %s", f[2]);
712 					goto badline;
713 				}
714 				score = scorebuf;
715 			}
716 			if(parsetime(f[3], &time) < 0){
717 				fprint(2, "%s:%d: bad time %s\n", name, line, f[3]);
718 				time = 1;
719 			}
720 			ctreemountfsys(c.ctree, f[1], time, score, image);
721 			continue;
722 		}
723 		if(strcmp(f[0], "allow") == 0 || strcmp(f[0], "deny") == 0){
724 			if(nf != 2){
725 				werrstr("syntax error: allow|deny ip[/mask]"); 
726 				goto badline;
727 			}
728 			c.ok = erealloc(c.ok, (c.nok+1)*sizeof(c.ok[0]));
729 			if(parseipandmask(f[1], c.ok[c.nok].ip, c.ok[c.nok].mask) < 0){
730 				werrstr("bad ip[/mask]: %s", f[1]);
731 				goto badline;
732 			}
733 			c.ok[c.nok].okay = (strcmp(f[0], "allow") == 0);
734 			c.nok++;
735 			continue;
736 		}
737 		werrstr("unknown verb '%s'", f[0]);
738 	badline:
739 		fprint(2, "%s:%d: %r\n", name, line);
740 	}
741 	Bterm(b);
742 
743 	sweepctree(c.ctree);
744 	free(cp->ok);
745 	*cp = c;
746 	return 0;
747 }
748 
749 int
750 ipokay(uchar *ip, ushort port)
751 {
752 	int i;
753 	uchar ipx[IPaddrlen];
754 	Ipokay *ok; 
755 	
756 	for(i=0; i<config.nok; i++){
757 		ok = &config.ok[i];
758 		maskip(ip, ok->mask, ipx);
759 if(0) fprint(2, "%I & %I = %I (== %I?)\n",
760 	ip, ok->mask, ipx, ok->ip);
761 		if(memcmp(ipx, ok->ip, IPaddrlen) == 0)
762 			return ok->okay;
763 	}
764 	if(config.nok == 0)	/* all is permitted */
765 		return 1;
766 	/* otherwise default is none allowed */
767 	return 0;
768 }
769 
770 Nfs3Status
771 cnodelookup(Ctree *t, Cnode **np, char *name)
772 {
773 	Cnode *n, *nn;
774 	
775 	n = *np;
776 	if(n->isblackhole)
777 		return Nfs3Ok;
778 	if((nn = cnodewalk(n, name, strlen(name), 0)) == nil){
779 		if(n->ismtpt || n->fsys){
780 			if((nn = cnodewalk(n, "", 0, 1)) == nil){
781 				nn = mkcnode(t, n, "", 0, (char*)n->handle, SHA1dlen);
782 				nn->isblackhole = 1;
783 			}
784 			nn->mark = 0;
785 		}
786 	}
787 	if(nn == nil)
788 		return Nfs3ErrNoEnt;
789 	*np = nn;
790 	return Nfs3Ok;
791 }
792 
793 Nfs3Status
794 cnodegetattr(Cnode *n, Nfs3Attr *attr)
795 {
796 	memset(attr, 0, sizeof *attr);
797 	if(n->read){
798 		attr->type = Nfs3FileReg;
799 		attr->mode = 0444;
800 		attr->size = 512;
801 		attr->nlink = 1;
802 	}else{
803 		attr->type = Nfs3FileDir;
804 		attr->mode = 0555;
805 		attr->size = 1024;
806 		attr->nlink = 10;
807 	}
808 	attr->fileid = *(u64int*)n->handle;
809 	attr->atime.sec = n->mtime;
810 	attr->mtime.sec = n->mtime;
811 	attr->ctime.sec = n->mtime;
812 	return Nfs3Ok;
813 }
814 
815 Nfs3Status
816 cnodereaddir(Cnode *n, u32int count, u64int cookie, uchar **pdata, u32int *pcount, u1int *peof)
817 {
818 	uchar *data, *p, *ep, *np;
819 	u64int c;
820 	Nfs3Entry ne;
821 	
822 	n = n->kidlist;
823 	c = cookie;
824 	for(; c && n; c--)
825 		n = n->nextsib;
826 	if(n == nil){
827 		*pdata = 0;
828 		*pcount = 0;
829 		*peof = 1;
830 		return Nfs3Ok;
831 	}
832 	
833 	data = emalloc(count);
834 	p = data;
835 	ep = data+count;
836 	while(n && p < ep){
837 		if(n->mark || n->name[0] == '+'){
838 			n = n->nextsib;
839 			++cookie;
840 			continue;
841 		}
842 		ne.name = n->name;
843 		ne.namelen = strlen(n->name);
844 		ne.cookie = ++cookie;
845 		ne.fileid = *(u64int*)n->handle;
846 		if(nfs3entrypack(p, ep, &np, &ne) < 0)
847 			break;
848 		p = np;
849 		n = n->nextsib;
850 	}
851 	*pdata = data;
852 	*pcount = p - data;
853 	*peof = n==nil;
854 	return Nfs3Ok;
855 }
856 
857 void
858 timerproc(void *v)
859 {
860 	for(;;){
861 		sleep(60*1000);
862 		sendp(timerchan, 0);
863 	}
864 }
865 
866 void
867 timerthread(void *v)
868 {
869 	for(;;){
870 		recvp(timerchan);
871 	/*	refreshconfig(); */
872 	}
873 }
874 
875 /*
876  * Actually serve the NFS requests.  Called from nfs3srv.c.  
877  * Each request runs in its own thread (coroutine).
878  *
879  * Decrypted handles have the form:
880  *
881  *	config[20] - SHA1 hash identifying a config tree node
882  *	glob[10] - SHA1 hash prefix identifying a glob state
883  *	fsyshandle[<=10] - disk file system handle (usually 4 bytes)
884  */
885  
886 /*
887  * A fid represents a point in the file tree.
888  * There are three components, all derived from the handle:
889  *
890  *	- config tree position (also used to find fsys)
891  *	- glob state for exclusions
892  *	- file system position
893  */
894 enum
895 {
896 	HAccess,
897 	HAttr,
898 	HWalk,
899 	HDotdot,
900 	HRead
901 };
902 typedef struct Fid Fid;
903 struct Fid
904 {
905 	Cnode *cnode;
906 	Fsys *fsys;
907 	Nfs3Handle fsyshandle;
908 };
909 
910 int
911 handlecmp(Nfs3Handle *h, Nfs3Handle *h1)
912 {
913 	if(h->len != h1->len)
914 		return h->len - h1->len;
915 	return memcmp(h->h, h1->h, h->len);
916 }
917 
918 Nfs3Status
919 handletofid(Nfs3Handle *eh, Fid *fid, int mode)
920 {
921 	int domount;
922 	Cnode *n;
923 	Disk *disk, *cdisk;
924 	Fsys *fsys;
925 	Nfs3Status ok;
926 	Nfs3Handle h2, *h, *fh;
927 	
928 	memset(fid, 0, sizeof *fid);
929 
930 	domount = 1;
931 	if(mode == HDotdot)
932 		domount = 0;
933 	/*
934 	 * Not necessary, but speeds up ls -l /dump/2005
935 	 * HAttr and HAccess must be handled the same way
936 	 * because both can be used to fetch attributes.
937 	 * Acting differently yields inconsistencies at mount points,
938 	 * and causes FreeBSD ls -l to fail.
939 	 */
940 	if(mode == HAttr || mode == HAccess)
941 		domount = 0;
942 
943 	/*
944 	 * Decrypt handle.
945 	 */
946 	h2 = *eh;
947 	h = &h2;
948 	if((ok = hdecrypt(h)) != Nfs3Ok)
949 		return ok;
950 	trace("handletofid: decrypted %.*lH\n", h->len, h->h);
951 	if(h->len < FsysHandleOffset)
952 		return Nfs3ErrBadHandle;
953 
954 	/*
955 	 * Find place in config tree.
956 	 */
957 	if((n = cnodebyhandle(config.ctree, h->h)) == nil)
958 		return Nfs3ErrStale;
959 	fid->cnode = n;
960 
961 	if(n->ismtpt && domount){
962 		/*
963 		 * Open fsys for mount point if needed.
964 		 */
965 		if(n->mfsys == nil){
966 			trace("handletofid: mounting %V/%s\n", n->fsysscore, n->fsysimage);
967 			if(n->fsysimage){
968 				if(strcmp(n->fsysimage, "/dev/null") == 0)
969 					return Nfs3ErrAcces;
970 				if((disk = diskopenfile(n->fsysimage)) == nil){
971 					fprint(2, "cannot open disk %s: %r\n", n->fsysimage);
972 					return Nfs3ErrIo;
973 				}
974 				if((cdisk = diskcache(disk, blocksize, 64)) == nil){
975 					fprint(2, "cannot cache disk %s: %r\n", n->fsysimage);
976 					diskclose(disk);
977 				}
978 				disk = cdisk;
979 			}else{
980 				if((disk = diskopenventi(vcache, n->fsysscore)) == nil){
981 					fprint(2, "cannot open venti disk %V: %r\n", n->fsysscore);
982 					return Nfs3ErrIo;
983 				}
984 			}
985 			if((fsys = fsysopen(disk)) == nil){
986 				fprint(2, "cannot open fsys on %V: %r\n", n->fsysscore);
987 				diskclose(disk);
988 				return Nfs3ErrIo;
989 			}
990 			n->mfsys = fsys;
991 			fsysroot(fsys, &n->mfsyshandle);
992 		}
993 		
994 		/*
995 		 * Use inner handle.
996 		 */
997 		fid->fsys = n->mfsys;
998 		fid->fsyshandle = n->mfsyshandle;
999 	}else{
1000 		/*
1001 		 * Use fsys handle from tree or from handle.
1002 		 * This assumes that fsyshandle was set by fidtohandle
1003 		 * earlier, so it's not okay to reuse handles (except the root)
1004 		 * across sessions.  The encryption above makes and 
1005 		 * enforces the same restriction, so this is okay.
1006 		 */
1007 		fid->fsys = n->fsys;
1008 		fh = &fid->fsyshandle;
1009 		if(n->isblackhole){
1010 			fh->len = h->len-FsysHandleOffset;
1011 			memmove(fh->h, h->h+FsysHandleOffset, fh->len);
1012 		}else
1013 			*fh = n->fsyshandle;
1014 		trace("handletofid: fsyshandle %.*lH\n", fh->len, fh->h);
1015 	}
1016 
1017 	/*
1018 	 * TO DO (maybe): some sort of path restriction here.
1019 	 */
1020 	trace("handletofid: cnode %s fsys %p fsyshandle %.*lH\n",
1021 		n->name, fid->fsys, fid->fsyshandle.len, fid->fsyshandle.h);
1022 	return Nfs3Ok;
1023 }
1024 
1025 void
1026 _fidtohandle(Fid *fid, Nfs3Handle *h)
1027 {
1028 	Cnode *n;
1029 	
1030 	n = fid->cnode;
1031 	/*
1032 	 * Record fsys handle in n, don't bother sending it to client
1033 	 * for black holes.
1034 	 */
1035 	n->fsys = fid->fsys;
1036 	if(!n->isblackhole){
1037 		n->fsyshandle = fid->fsyshandle;
1038 		fid->fsyshandle.len = 0;
1039 	}
1040 	memmove(h->h, n->handle, CnodeHandleSize);
1041 	memmove(h->h+FsysHandleOffset, fid->fsyshandle.h, fid->fsyshandle.len);
1042 	h->len = FsysHandleOffset+fid->fsyshandle.len;
1043 }
1044 
1045 void
1046 fidtohandle(Fid *fid, Nfs3Handle *h)
1047 {
1048 	_fidtohandle(fid, h);
1049 	hencrypt(h);
1050 }
1051 
1052 void
1053 setrootfid(void)
1054 {
1055 	Fid fid;
1056 	
1057 	memset(&fid, 0, sizeof fid);
1058 	fid.cnode = config.ctree->root;
1059 	_fidtohandle(&fid, &root);
1060 }
1061 
1062 void
1063 fsgetroot(Nfs3Handle *h)
1064 {
1065 	*h = root;
1066 	hencrypt(h);
1067 }
1068 
1069 Nfs3Status
1070 fsgetattr(SunAuthUnix *au, Nfs3Handle *h, Nfs3Attr *attr)
1071 {
1072 	Fid fid;
1073 	Nfs3Status ok;
1074 
1075 	trace("getattr %.*lH\n", h->len, h->h);
1076 	if((ok = handletofid(h, &fid, HAttr)) != Nfs3Ok)
1077 		return ok;
1078 	if(fid.fsys)
1079 		return fsysgetattr(fid.fsys, au, &fid.fsyshandle, attr);
1080 	else
1081 		return cnodegetattr(fid.cnode, attr);
1082 }
1083 
1084 /*
1085  * Lookup is always the hard part.
1086  */
1087 Nfs3Status
1088 fslookup(SunAuthUnix *au, Nfs3Handle *h, char *name, Nfs3Handle *nh)
1089 {
1090 	Fid fid;
1091 	Cnode *n;
1092 	Nfs3Status ok;
1093 	Nfs3Handle xh;
1094 	int mode;
1095 	
1096 	trace("lookup %.*lH %s\n", h->len, h->h, name);
1097 
1098 	mode = HWalk;
1099 	if(strcmp(name, "..") == 0 || strcmp(name, ".") == 0)
1100 		mode = HDotdot;
1101 	if((ok = handletofid(h, &fid, mode)) != Nfs3Ok){
1102 		nfs3errstr(ok);
1103 		trace("lookup: handletofid %r\n");
1104 		return ok;
1105 	}
1106 	
1107 	if(strcmp(name, ".") == 0){
1108 		fidtohandle(&fid, nh);
1109 		return Nfs3Ok;
1110 	}
1111 
1112 	/*
1113 	 * Walk down file system and cnode simultaneously.
1114 	 * If dotdot and file system doesn't move, need to walk
1115 	 * up cnode.  Save the corresponding fsys handles in
1116 	 * the cnode as we walk down so that we'll have them 
1117 	 * for dotdotting back up.
1118 	 */
1119 	n = fid.cnode;
1120 	if(mode == HWalk){
1121 		/*
1122 		 * Walk down config tree and file system simultaneously.
1123 		 */
1124 		if((ok = cnodelookup(config.ctree, &n, name)) != Nfs3Ok){
1125 			nfs3errstr(ok);
1126 			trace("lookup: cnodelookup: %r\n");
1127 			return ok;
1128 		}
1129 		fid.cnode = n;
1130 		if(fid.fsys){
1131 			if((ok = fsyslookup(fid.fsys, au, &fid.fsyshandle, name, &xh)) != Nfs3Ok){
1132 				nfs3errstr(ok);
1133 				trace("lookup: fsyslookup: %r\n");
1134 				return ok;
1135 			}
1136 			fid.fsyshandle = xh;
1137 		}		
1138 	}else{
1139 		/*
1140 		 * Walking dotdot.  Ick.
1141 		 */
1142 		trace("lookup dotdot fsys=%p\n", fid.fsys);
1143 		if(fid.fsys){
1144 			/*
1145 			 * Walk up file system, then try up config tree.
1146 			 */
1147 			if((ok = fsyslookup(fid.fsys, au, &fid.fsyshandle, "..", &xh)) != Nfs3Ok){
1148 				nfs3errstr(ok);
1149 				trace("lookup fsyslookup: %r\n");
1150 				return ok;
1151 			}
1152 			fid.fsyshandle = xh;
1153 
1154 			/*
1155 			 * Usually just go to n->parent.
1156 			 * 
1157 			 * If we're in a subtree of the mounted file system that
1158 			 * isn't represented explicitly by the config tree (instead
1159 			 * the black hole node represents the entire file tree),
1160 			 * then we only go to n->parent when we've dotdotted back
1161 			 * to the right handle.
1162 			 */
1163 			if(n->parent == nil)
1164 				trace("lookup dotdot: no parent\n");
1165 			else{
1166 				trace("lookup dotdot: parent %.*lH, have %.*lH\n",
1167 					n->parent->fsyshandle.len, n->parent->fsyshandle.h,
1168 					xh.len, xh.h);
1169 			}
1170 			
1171 			if(n->isblackhole){
1172 				if(handlecmp(&n->parent->mfsyshandle, &xh) == 0)
1173 					n = n->parent;
1174 			}else{
1175 				if(n->parent)
1176 					n = n->parent;
1177 			}
1178 		}else{
1179 			/*
1180 			 * No file system, just walk up.
1181 			 */
1182 			if(n->parent)
1183 				n = n->parent;
1184 		}
1185 		fid.fsys = n->fsys;
1186 		if(!n->isblackhole)
1187 			fid.fsyshandle = n->fsyshandle;
1188 		fid.cnode = n;
1189 	}
1190 	fidtohandle(&fid, nh);
1191 	return Nfs3Ok;
1192 }
1193 
1194 Nfs3Status
1195 fsaccess(SunAuthUnix *au, Nfs3Handle *h, u32int want, u32int *got, Nfs3Attr *attr)
1196 {
1197 	Fid fid;
1198 	Nfs3Status ok;
1199 	
1200 	trace("access %.*lH 0x%ux\n", h->len, h->h, want);
1201 	if((ok = handletofid(h, &fid, HAccess)) != Nfs3Ok)
1202 		return ok;
1203 	if(fid.fsys)
1204 		return fsysaccess(fid.fsys, au, &fid.fsyshandle, want, got, attr);
1205 	*got = want & (Nfs3AccessRead|Nfs3AccessLookup|Nfs3AccessExecute);
1206 	return cnodegetattr(fid.cnode, attr);
1207 }
1208 
1209 Nfs3Status
1210 fsreadlink(SunAuthUnix *au, Nfs3Handle *h, char **link)
1211 {
1212 	Fid fid;
1213 	Nfs3Status ok;
1214 	
1215 	trace("readlink %.*lH\n", h->len, h->h);
1216 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1217 		return ok;
1218 	if(fid.fsys)
1219 		return fsysreadlink(fid.fsys, au, &fid.fsyshandle, link);
1220 	*link = 0;
1221 	return Nfs3ErrNotSupp;
1222 }
1223 
1224 Nfs3Status
1225 fsreadfile(SunAuthUnix *au, Nfs3Handle *h, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1226 {
1227 	Fid fid;
1228 	Nfs3Status ok;
1229 	
1230 	trace("readfile %.*lH\n", h->len, h->h);
1231 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1232 		return ok;
1233 	if(fid.cnode->read)
1234 		return fid.cnode->read(fid.cnode, count, offset, data, pcount, peof);
1235 	if(fid.fsys)
1236 		return fsysreadfile(fid.fsys, au, &fid.fsyshandle, count, offset, data, pcount, peof);
1237 	return Nfs3ErrNotSupp;
1238 }
1239 
1240 Nfs3Status
1241 fsreaddir(SunAuthUnix *au, Nfs3Handle *h, u32int len, u64int cookie, uchar **pdata, u32int *pcount, u1int *peof)
1242 {
1243 	Fid fid;
1244 	Nfs3Status ok;
1245 
1246 	trace("readdir %.*lH\n", h->len, h->h);	
1247 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1248 		return ok;
1249 	if(fid.fsys)
1250 		return fsysreaddir(fid.fsys, au, &fid.fsyshandle, len, cookie, pdata, pcount, peof);
1251 	return cnodereaddir(fid.cnode, len, cookie, pdata, pcount, peof);
1252 }
1253 
1254 Nfs3Status
1255 logread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1256 {
1257 	*pcount = 0;
1258 	*peof = 1;
1259 	return Nfs3Ok;
1260 }
1261 
1262 Nfs3Status
1263 refreshdiskread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1264 {
1265 	char buf[128];
1266 	
1267 	if(offset != 0){
1268 		*pcount = 0;
1269 		*peof = 1;
1270 		return Nfs3Ok;
1271 	}
1272 	if(refreshdisk() < 0)
1273 		snprint(buf, sizeof buf, "refreshdisk: %r\n");
1274 	else
1275 		strcpy(buf, "ok\n");
1276 	*data = emalloc(strlen(buf));
1277 	strcpy((char*)*data, buf);
1278 	*pcount = strlen(buf);
1279 	*peof = 1;
1280 	return Nfs3Ok;
1281 }
1282 
1283 Nfs3Status
1284 refreshconfigread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1285 {
1286 	char buf[128];
1287 	
1288 	if(offset != 0){
1289 		*pcount = 0;
1290 		*peof = 1;
1291 		return Nfs3Ok;
1292 	}
1293 	if(readconfigfile(&config) < 0)
1294 		snprint(buf, sizeof buf, "readconfig: %r\n");
1295 	else
1296 		strcpy(buf, "ok\n");
1297 	*data = emalloc(strlen(buf));
1298 	strcpy((char*)*data, buf);
1299 	*pcount = strlen(buf);
1300 	*peof = 1;
1301 	return Nfs3Ok;
1302 }
1303